惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Y
Y Combinator Blog
Jina AI
Jina AI
雷峰网
雷峰网
有赞技术团队
有赞技术团队
WordPress大学
WordPress大学
美团技术团队
V
V2EX
酷 壳 – CoolShell
酷 壳 – CoolShell
小众软件
小众软件
博客园 - Franky
博客园 - 三生石上(FineUI控件)
月光博客
月光博客
博客园 - 叶小钗
大猫的无限游戏
大猫的无限游戏
爱范儿
爱范儿
Hugging Face - Blog
Hugging Face - Blog
宝玉的分享
宝玉的分享
Last Week in AI
Last Week in AI
Apple Machine Learning Research
Apple Machine Learning Research
量子位
IT之家
IT之家
人人都是产品经理
人人都是产品经理
博客园_首页
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com

Hacker News: Show HN

PurrrrrFocus: Pomodoro Timer App - App Store Workflow Engine — Multi-Step Orchestration for Bun RapidPhoto: Pro Photo Editor App - App Store GitHub - DheerG/swarms: Achieve extraordinary results with claude code across a variety of tasks SPICE simulation → oscilloscope → verification with Claude Code — Lucas Gerads Show HN: VCoding – A 5 MB native Windows IDE with no dynamic dependencies Show HN: LLMs don't hallucinate because they're bad at math, it's the format GitHub - Agent-FM/agentfm-core: AgentFM is a peer-to-peer network that turns everyday computers into a decentralized AI supercomputer. AgentFM lets you run massive AI workloads directly across a global mesh of idle CPUs and GPUs. Show HN: Tracking Top US Science Olympiad Alumni over Last 25 Years GitHub - Potarix/agent-hub: One place to talk to all your agents Show HN: Runtime security for AI agents(injection,tool abuse, data exfiltration) GitHub - dubeyKartikay/lazyspotify: Terminal Spotify client for macOS and Linux GitHub - the-banana-tool/king-louie: Easy to use GUI Personal AI Assistant. Win/Linux/Mac. Show HN I made my vacation rental bookable by AI agents–no Airbnb, 0% commission GitHub - basteez/jsf-autoreload: maven plugin to enable hot reload on jsf projects uvm32/hosts/host-gdbstub at main · ringtailsoftware/uvm32 GitHub - labsai/EDDI: Config-driven engine that turns JSON into production-grade AI agents. Multi-agent orchestration, 12+ LLM providers, MCP/A2A protocols, RAG, persistent memory, and enterprise compliance (EU AI Act, GDPR, HIPAA). Built on Quarkus. GitHub - glitchnsec/fortyone-oss: AI Executive Assistant Platform Quickstart | Alien GitHub - muxshed/shed: One stream in, or many. Every destination, simultaneously. No cloud middleman, no per-channel fees, no limits. GitHub - ocrbase-hq/ocrbase: 📄 PDF/IMG ->.MD/JSON Document OCR API for PaddleOCR and GLMOCR. Self-hostable. GitHub - impactjo/home-memory: MCP server that lets your AI assistant remember everything about your home. GitHub - Sets88/dbcls: DbCls is a powerful terminal database client that supports various databases GitHub - neptun2000/heor-agent-mcp GitHub - SeanFDZ/macmind: Single-layer transformer in HyperTalk for the classic Macintosh RollQuation: Math Puzzles - Apps on Google Play GitHub - dropbox/witchcraft Show HN: Agent-cache – Multi-tier LLM/tool/session caching for Valkey and Redis GitHub - opentalon/opentalon: OpenTalon is an open-source platform built from the ground up in Go as a robust alternative to OpenClaw LinkedIn™ 职位抓取工具 - Chrome 应用商店
GitHub - elliotgao2/handsets: A high-performance Android ...
gaojiuli · 2026-05-21 · via Hacker News: Show HN

Handsets — a high-performance Android control CLI, built for agents and humans.

The device control plane you wished adb was.


Install

curl -fsSL https://raw.githubusercontent.com/elliotgao2/handsets/main/install.sh | bash

macOS and Linux. Pin a version with HANDSETS_VERSION=v0.1.0 …. Then hs use against a connected device.

Build from source
git clone https://github.com/elliotgao2/handsets && cd handsets
./build.sh                                                          # daemon jar
cargo build --release --manifest-path handsets-cli/Cargo.toml       # `hs`
cargo build --release --manifest-path handsets-viewer/Cargo.toml    # `hs see` GUI (macOS)
ln -s "$PWD/handsets-cli/target/release/hs" /usr/local/bin/hs

At a glance

$ hs                                           # list devices
$ hs use                                       # connect, start daemon, mirror state
$ hs info                                      # neofetch-style snapshot (2 ms)
$ hs see x.jpg                                 # screenshot — 100× faster than adb
$ hs ui -i                                     # flat list of tappable nodes (LLM-friendly)
$ hs find 'TextView[text~=Login]'              # CSS-like selector over the live tree
$ hs open com.foo/.MainActivity && hs wait com.foo
$ hs type "EditText" "user@example.com"        # ACTION_SET_TEXT, no virtual keyboard
$ hs do <<'EOF'                                # persistent shell — many cmds, one socket
  tap_and_dump x=540 y=1500 idle_ms=200
EOF

hs ui -i — the agent-friendly UI dump

uiautomator2's d.dump_hierarchy() and Appium's driver.page_source hand you the full XML — hundreds of KB per screen, mostly layout noise. hs ui -i filters and flattens the same tree to only the nodes a human or LLM can actually act on, usually 10–100× less text:

@(54,160)    click             ImageButton                desc="Back"
@(540,360)                     TextView    #title         "Sign in to your account"
@(540,540)   click             EditText    #email         desc="Email"
@(540,640)   click,password    EditText    #password      desc="Password"
@(540,760)   check             CheckBox                   "Remember me"
@(540,860)   click             Button      #continue      "Continue"
@(540,960)   click             TextView                   "Forgot password?"

Four columns: center coords / behaviour tags / class+id / text or desc. Drop into an LLM, get back coords, hand them to hs tap X Y — done.

Verbs

Devices

hs                                  list attached devices
hs use   [SERIAL]                   connect; auto-spawns the state mirror
hs drop  [SERIAL] [--keep-jar]      disconnect

Inspect

hs info                             neofetch-style snapshot (2 ms, from local cache)
hs see                              live viewer (Metal + VideoToolbox H.264)
hs see   foo.{jpg,png,xml,json}     capture, format by extension
hs ui    [-i|--json|--xml] [--all]  UI tree dump — `-i` returns a flat,
                                      LLM-friendly table of tappable nodes
                                      (coords + tags + class + label)
hs find  SELECTOR                   CSS-like:  Tag[attr=val]:flag, comma = OR
hs show  [top | PKG]                device state | top activity | package info
hs apps  [--3rd]                    installed packages

Activity

hs open      COMPONENT              start activity
hs close     PKG                    force-stop
hs install   APK [APK …]            streamed PackageInstaller, multi-APK
hs uninstall PKG

Input

hs tap   "Login" | X Y              text-lookup or coords
hs type  TEXT                       KeyEvents to the focused field
hs type  SELECTOR TEXT              ACTION_SET_TEXT — atomic, bypasses the IME
hs go    back | home | recents | …  key events
hs swipe left|right|up|down [DUR_MS]    80% screen swipe (daemon picks coords)
hs swipe X1 Y1 X2 Y2 [DUR_MS]

Sync

hs wait  idle [Nms] | TEXT | PKG | Nms      event-driven, no polling
hs cp    device:src dst | src device:dst    scp-style file transfer

System

hs prop     [KEY [VAL]]             bare = list all; KEY = get; KEY VAL = set
hs settings [NS [KEY [VAL]]]        bare = list all 3; NS = list one;
                                      NS KEY = get; NS KEY VAL = set

Diagnostics

hs logs   [--tail N | --follow]     logcat (default last 100)
hs events                           lifecycle stream (am monitor)

Shell

hs shell                            interactive REPL (history, built-ins,
                                      unknown verbs fall through to /system/bin/sh)
hs do     [WIRE]                    same REPL, or one-shot raw wire

vs uiautomator2, Appium

Both wrap UIAutomator and pay for the framework. Handsets runs a hand-rolled daemon directly under app_process and talks to binders by reflection, so the wire is tighter and the dump format is built for agents rather than reporters.

Handsets uiautomator2 Appium
Wire TCP long socket, length-prefixed binary HTTP/JSON via atx-agent WebDriver over HTTP
On-device install push 1 jar (~few hundred KB) 2 apks + atx-agent driver apk + Node server
Daemon start < 200 ms via app_process, no UIAutomator framework UIAutomator instrumentation each session UIAutomator + WebDriver bridge
State reads µs from host-mirrored file (hs info / hs show) ms per round-trip ms+ per round-trip
UI dump for agents hs ui -i flat, ~10× fewer tokens full XML full XML
Selector CSS-like Tag[attr=val][attr~=sub]:flag d(text=…, className=…) chained Selenium strategies
Bound to any language via subprocess Python only multi-lang via WebDriver
Best at LLM agents, ad-hoc scripts, high-freq small ops Python device scraping cross-platform CI suites

Honest tradeoff: uiautomator2 and Appium ship with recorders, IDE integrations, pytest runners, HTML reporting. Handsets is a lean CLI. For pytest-style UI regression with reports they're still the smoother path. Handsets is built for the case where you only care about single-call latency and shell composition.

Docs